Re: How to track number of connections and hosts to Postgres cluster

On Wed, Aug 24, 2011 at 9:33 AM, Greg Smith <greg(at)2ndquadrant(dot)com> wrote:

> On 08/24/2011 07:07 AM, Venkat Balaji wrote:
>
>> But, if put log_connections to on and log_disconnections to on wouldn't
>> the Postgres be logging in lot of data ?
>> Will this not be IO intensive ? I understand that this is the best way,
>> but, would want to know if there is an other way to reduce IO ( may be
>> through queries to catalog tables ).
>>
>>
> Your requirements include: " I need all the host IPs making a connection
> to Postgres Cluster (even for a fraction of second)."
>
> The only way to do this is to log every connection. Any other approach for
> grabbing the data, such as looking at pg_stat_activity, will sometimes miss
> one.
>
> If you're willing to lose a connection sometimes, a cron job that polls
> pg_stat_activity and saves a summary of what it finds will normally use less
> resources. But connections that start and end between runs will be missed.
>
>
I suppose you could use tcpdump on a separate system with a mirrored switch
port and have it log TCP SYN and FIN packets on port 5432 to your database
server only. Keeps all I/O off your database server.

tcpdump -w port5423.log -n "tcp and port 5432 and tcp[tcpflags] &
(tcp-syn|tcp-fin) != 0 and host IP"

HTH.

Greg